https://www.youngju.dev/blog 천천히 올바르게. AI Researcher & DevOps Engineer Youngju's blog. GPU/CUDA, LLM, MLOps, Kubernetes AI workloads, and data engineering — plus mindset essays on confidence, routines, health, and sport psychology. ko fjvbn2003@gmail.com (Youngju Kim) fjvbn2003@gmail.com (Youngju Kim) Fri, 14 Aug 2026 00:00:00 GMT https://www.youngju.dev/blog/observability/2026-08-14-langfuse-querying-and-evals.en https://www.youngju.dev/blog/observability/2026-08-14-langfuse-querying-and-evals.en Collecting traces and getting answers out of them are different jobs. This post lays out which metric axes Langfuse offers and which dimensions you have to slice them by, then walks the Metrics API v2 query structure as an actual request. It covers the v4 change that removed the traces view in favour of an observations-centric model, the four places a score attaches, what moving evaluators from trace level down to observation level makes possible, and the configuration needed to enable code evaluators when self-hosting. It closes with an order for designing dashboards that starts from questions rather than charts. Final post in the Langfuse and ClickHouse LLM tracing series. Fri, 14 Aug 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) observabilitylangfusellm-evaluationmetrics-apidashboardllmops https://www.youngju.dev/blog/observability/2026-08-14-langfuse-querying-and-evals.ja https://www.youngju.dev/blog/observability/2026-08-14-langfuse-querying-and-evals.ja トレースを集めることと、そのトレースから答えを得ることは別の仕事です。本記事はLangfuseが提供する指標の軸は何で、それをどの次元で切って見るべきかを整理したうえで、メトリクスAPI v2の問い合わせ構造を実際のリクエストの形で確認します。v4でtracesビューが姿を消しobservations中心に再編された変化、scoreがtrace・observation・session・データセット実行の四か所に付く仕組み、評価器がtraceレベルからobservationレベルへ降りたことが何を可能にするか、そしてコード評価器をセルフホストで有効にする際に必要な設定まで扱います。最後に、ダッシュボードを画面ではなく問いから設計する順序を提案します。Langfuse + ClickHouse LLMトレーシングシリーズの最終回です。 Fri, 14 Aug 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) observabilitylangfusellm-evaluationmetrics-apidashboardllmops https://www.youngju.dev/blog/observability/2026-08-14-langfuse-querying-and-evals https://www.youngju.dev/blog/observability/2026-08-14-langfuse-querying-and-evals 트레이스를 모으는 것과 그 트레이스로 답을 얻는 것은 다른 일입니다. 이 글은 Langfuse가 제공하는 지표 축이 무엇이고 그것을 어떤 차원으로 쪼개 봐야 하는지 정리한 뒤, 메트릭 API v2의 질의 구조를 실제 요청 형태로 확인합니다. v4에서 traces 뷰가 사라지고 observations 중심으로 재편된 변화, score가 trace·observation·session·데이터셋 실행 네 곳에 붙는 방식, 평가자가 trace 수준에서 observation 수준으로 내려온 것이 무엇을 가능하게 하는지, 그리고 코드 평가자를 자체 호스팅에서 켤 때 필요한 설정까지 다룹니다. 마지막으로 대시보드를 화면이 아니라 질문에서 시작해 설계하는 순서를 제안합니다. Langfuse + ClickHouse LLM 트레이싱 시리즈의 마지막 글입니다. Fri, 14 Aug 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) observabilitylangfusellm-evaluationmetrics-apidashboardllmops